基于地理知识图谱的智能问答系统设计与实现  

Design and Implementation of an Intelligent Question and Answer System based on Geographical Knowledge Graph

作　　者：巴图 尹川 张彬钰 杜明义 李志伟 Ba Tu;Yin Chuan;Zhang Binyu;Du Mingyi;Li Zhiwei(School of Geomatics and Urban Spatial Informatics,Beijing University of Civil Engineering and Architecture,Beijing,China)

机构地区：[1]北京建筑大学测绘与城市空间信息学院,北京

出　　处：《科学技术创新》2023年第7期29-32,共4页Scientific and Technological Innovation

基　　金：北京市教育委员会科学研究计划项目资助(编号:KM202110016003)。

摘　　要：用知识图谱作为数据源,为问答系统提供高质量知识。本研究基于地理领域知识图谱,设计了地理知识的智能问答系统。(1)通过Doc2vec+k-means聚类算法生成问题模板;(2)利用多项式朴素贝叶斯分类,根据模板的语义及问题中的实体识别用户的问题意图;(3)利用空间关系及语义约束关系实现知识推理,完成答案检索。本研究抽取部分问题作为问题进行实验,将问题分为空间类问题和数据类问题,统计得到模板匹配的平均准确率为97.0%,召回率为96.7%,F1为96.8%;最终答案生成的平均准确率为84.5%,能为地理服务智能化和信息化提供支持。s:We use the knowledge graph as a data source to provide high-quality knowledge for the question and answer system.In this paper,we design an intelligent question and answer system for geographic knowledge based on geographic domain knowledge graphs.(1)generating question templates by Doc2vec+k-means clustering algorithm;(2)using polynomial plain Bayesian classification to identify users'question intentions based on the semantics of the templates and the entities in the questions;(3)using spatial relations and semantic constraint relations to realize knowledge inference and complete answer retrieval.In this paper,some questions are extracted as problems for experiments,and the questions are divided into spatial and data class questions,and the average accuracy of template matching is 97.0%,the recall rate is 96.7%,and the F1 is 96.8%;the average accuracy of final answer generation is 84.5%,which can provide support for geographic service intelligence and informationization.

关 键 词：地理知识图谱 智能问答 空间关系 语义约束 知识推理 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]